Assessing the impact of data augmentation and a combination of CNNs on leukemia classification

doi:10.1016/j.ins.2022.07.059

Information Sciences

Volume 609, September 2022, Pages 1010-1029

https://doi.org/10.1016/j.ins.2022.07.059 Get rights and content

Highlights

•
Deep learning for detecting types of leukemia in blood smear images.
•
Seven CNNs were applied in five scenarios for different types of leukemia.
•
Data Augmentation was applied and analyzed to prove the performance improvement.
•
Multilevel and Ensemble improve the individual CNN’s performance.
•
Multilevel and Ensemble performances are similar.

Abstract

An accurate early-stage leukemia diagnosis plays a critical role in treating and saving patients’ lives. The two primary forms of leukemia are acute and chronic leukemia, which is subdivided into myeloid and lymphoid leukemia. Deep learning models have been increasingly used in computer-aided medical diagnosis (CAD) systems developed to detect leukemia. This article assesses the impact of widely applied techniques, mainly data augmentation and multilevel and ensemble configurations, in deep learning-based CAD systems. Our assessment included five scenarios: three binary classification problems and two multiclass classification problems. The evaluation was performed using 3,536 images from 18 datasets, and it was possible to conclude that data augmentation techniques improve the performance of convolutional neural networks (CNNs). Furthermore, there is an improvement in the classification results using a combination of CNNs. For the binary problems, the performance of the ensemble configuration was superior to that of the multilevel configuration. However, the results were statistically similar in multiclass scenarios. The results were promising, with accuracies of 94.73% and 94.59% obtained using multilevel and ensemble configurations in a scenario with four classes. The combination of methods helps to reduce the error or variance of the predictions, which improves the accuracy of the used deep learning-based model.

Introduction

Bone marrow occupies the bone cavity, where blood cells are produced. It contains the cells that give rise to red blood cells, known as erythrocytes, platelets, and white blood cells, also known as leukocytes. The latter cells actively participate in the human immune system and help it to defend the body against invaders. Progenitor cells in the marrow, also known as stem cells or precursor cells, produce an average of 100 million leukocytes per day. These leukocytes help the body to combat and eliminate microorganisms and chemical structures that are strangers to it through their capture, i.e., phagocytosis or through the production of antibodies. One of the diseases that affect the functioning of the bone marrow is leukemia [45].

Leukemia is a malignant disease of the white blood cells, usually of unknown origin. Its main characteristic is the accumulation of diseased cells in the bone marrow, which replace normal blood cells. A blood cell that has not yet reached maturity undergoes a genetic mutation that turns it into a cancer cell in leukemia. This abnormal cell does not operate properly, and it multiplies faster and has a shorter lifespan than of normal cells. Hence, the abnormal cancer cells replace healthy blood cells in the bone marrow.

The American Cancer Society (ACS)¹ estimated that there would be 60,650 new cases of leukemia in 2022, with approximately 24,000 deaths; in particular, there would be 35,810 male cases and 24,840 female cases, leading to 14,020 male deaths and 9,980 female deaths.

The types of leukemia can be classified according to the worsening speed of the disease. Hence, the condition can be of the chronic type, which usually gets worse slowly, or of the acute type, which usually gets worse quickly. The types of leukemia can also be classified based on the kind of white blood cells they affect: lymphoid or myeloid cells. Thus, the main types of leukemia are acute lymphoblastic leukemia (ALL), acute myeloid leukemia (AML), chronic myeloid leukemia (CML), and chronic lymphocytic leukemia (CLL). Acute leukemia affects mainly children, and chronic leukemia tends to affect adults and the elderly [45].

Each type of leukemia has an appropriate treatment; therefore, a diagnosis in the early stage of the disease is demanded to provide the proper treatment successfully. On the other hand, the main treatments for more advanced disease phases aim to destroy the leukemic cells so that the bone marrow returns to produce normal cells. Fig. 1 shows examples of the blood slide images used in the experiments of the current study, mainly ALL, AML, chronic leukemia, and healthy blood slides (HBS).

Deep learning models have been increasingly used in computer-aided medical diagnosis (CAD) systems. In particular, convolutional neural networks (CNNs) can learn hierarchical representations, from more general features in the first convolutional layers, to more semantic features in the last few layers. Currently, CNNs are one of the most effective techniques used in medical imaging-based diagnosis [22]. Researchers have been seeking to increase the generalizability of CNNs, particularly based on techniques of data augmentation and the combination of CNNs in ensemble and multilevel configurations.

In this study, techniques that are widely used in CNN based CAD systems were evaluated, mainly data augmentation and ensemble and multilevel configurations. Therefore, seven CNNs were studied using different techniques of data augmentation and ensemble and multilevel configurations. According to five leukemia classification scenarios, the analysis was performed using 3,536 images from 18 heterogeneous datasets. Three of these scenarios are binary classification problems: leukemia vs. HBS, ALL vs. HBS, and AML vs. HBS. The other two scenarios are multiclass classification problems: ALL vs. AML vs. HBS and ALL vs. AML vs. HBS vs. other types.

The main contributions of this article are the following: the identifications of the datasets that are widely used for leukemia classification, the introduction of five scenarios for the classification of different types of leukemia, the evaluation of the performance achieved by various CNN-based models on leukemia classification, the assessment of the impact of multiple data augmentation techniques on the classification performance, and assessment and comparison of the improvements achieved by multilevel and ensemble model configurations.

This article is organized as follows. Section 2 presents related work. Section 3 describes the used materials and methods, such as the used datasets, the employed techniques of data augmentation, the evaluated network architectures, the used ensemble and multilevel configurations, and, finally, the adopted evaluation metrics. Sections 4 Experiments and results, 5 Discussion present the achieved results and a comparison of them against the ones of previous works found in the literature, respectively. Finally, the conclusions and possibilities for future work are pointed out in Section 6.

Section snippets

Related work

This section presents studies that have been developed for leukemia detection. Taking into account the applied methodology, we identified traditional methods [40], [29], [14], [32], [26] and methods based on deep learning [43], [48], [27], [1]. Traditional methods comprise several steps, such as image pre-processing, segmentation, feature extraction, and classification. On the other hand, procedures based on deep learning usually apply CNNs. This kind of procedures aims to design and build a

Materials and methods

This study aimed to evaluate the influence of using data augmentation and combinations of CNNs on the detection of leukemia types in blood slide images. The identification of leukemia types in images is a challenging issue. Here, five leukemia classification problems were addressed, mainly three binary classification and two multiclass classification problems: 1) leukemia vs. HBS, 2) ALL vs. HBS, 3) AML vs. HBS, 4) ALL vs. AML vs. HBS, and 5) ALL vs. AML vs. HBS vs. other types. Public image

Experiments and results

The dataset used to study the five scenarios under evaluation is composed of the following images: 1,434 images of healthy slides, 881 images of ALL, 978 images of AML and 243 images of “other types” of leukemia. K-fold cross-validation with the value of k equal to 5 was applied in the evaluated experiments, which were performed on a PC with a 3.6 GHz Intel®Xeon™sprocessor with 24 GB of RAM and an NVIDIA TITAN XP 12 GB graphics card.

The influence of the use of data augmentation on the

Discussion

Table 11 allows a comparison among related state-of-the-art methods regarding the addressed classification problem, used number of datasets, used number of images and achieved accuracy.

The obtained results suggest that even using general-purpose CNNs, by choosing suitable techniques of data augmentation and a appropriate combination of CNNs, results that are competitive against the state-of-the-art methods can be achieved.

To make a more reliable comparison, Table 11 is organized according to

Conclusion

In this study, techniques that can be integrated into computer-aided diagnostic systems in order to detect different types of leukemia, mainly ALL, AML, and other types, in addition to healthy slides, were evaluated. Several experiments were We performed. First, tests were performed according to five scenarios and the effectiveness of using techniques of data augmentation was analyzed. Then, a comparison among techniques of data augmentation for the ALL vs. AML vs. HBS vs. other types

CRediT authorship contribution statement

Maila L. Claro: Methodology, Software, Investigation, Writing – original draft. Rodrigo M.S. de Veras: Supervision, Writing – original draft. Andre M. Santana: Supervision, Writing – review & editing. Luis Henrique S. Vogado: Software, Writing – review & editing. Geraldo Braz Junior: Writing – review & editing. Fatima N.S. de Medeiros: Writing – review & editing. Joao Manuel R.S. Tavares: Writing – review & editing.

Declaration of Competing Interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Acknowledgment

This study was partially founded by the “Coordenação de Aperfeiçoamento de Pessoal de Nivel Superior” (CAPES) - Finance Code 001, “Fundação de Amparo a Pesquisa do Piaui ”(FAPEPI), and “Conselho Nacional de Desenvolvimento Cientifico e Tecnologico” (CNPQ), in Brazil. The authors gratefully acknowledge the support of NVIDIA Corporation with the donation of the Titan Xp GPU used in this study.

References (50)

A. Abhishek et al.
Automated classification of acute leukemia on a heterogeneous dataset using machine learning and deep learning techniques
Biomed. Signal Process. Control
(2022)
K. Anilkumar et al.
A survey on image segmentation of blood and bone marrow smear images with emphasis to automated detection of leukemia
Biocybern. Biomed. Eng.
(2020)
P.K. Das et al.
An efficient deep convolutional neural network based detection and classification of acute lymphoblastic leukemia
Expert Syst. Appl.
(2021)
S. Gehlot et al.
Sdct-auxnetθ: Dct augmented stain deconvolutional cnn with auxiliary classifier for cancer diagnosis
Med. Image Anal.
(2020)
J. Laosai et al.
Classification of acute leukemia using medical-knowledge-based morphology and cd marker
Biomed. Signal Process. Control
(2018)
J. Rawat et al.
HS, B., Virmani, J., Devgun, J.S. Computer assisted classification framework for prediction of acute lymphoblastic and acute myeloblastic leukemia
Biocybern. Biomed. Eng.
(2017)
S.H. Rezatofighi et al.
Automatic recognition of five types of white blood cells in peripheral blood
Comput. Med. Imaging Graph.
(2011)
L.H.S. Vogado et al.
Leukemia diagnosis in blood slides using transfer learning in cnns and SVM for classification
Eng. Appl. Artif. Intell.
(2018)
X. Zheng et al.
Fast and robust segmentation of white blood cell images by self-supervised learning
Micron
(2018)
N. Ahmed et al.
Identification of leukemia subtypes from microscopic images using convolutional neural network
Diagnostics
(2019)

ASH, Accessed in September 2020. Ash image bank: American society of hematology....

M. Belkin et al.

Reconciling modern machine-learning practice and the classical bias–variance trade-off

Proc. Nat. Acad. Sci.

(2019)

N. Bibi et al.

Iomt-based automated detection and classification of leukemia using deep learning

J. Healthcare Eng.

(2020)

J. Böhm

Pathologie-websites im world wide web

Der Pathologe

(2008)

F. Chollet

Xception: Deep learning with depthwise separable convolutions

M. Claro et al.

Convolution neural network models for acute leukemia diagnosis

B.V. Dasarathy et al.

A composite classifier system design: Concepts and methodology

Proc. IEEE

(1979)

S. Dasariraju et al.

Detection and classification of immature leukocytes for diagnosis of acute myeloid leukemia using random forest algorithm

Bioengineering

(2020)

D. Goutam et al.

Classification of acute myelogenous leukemia in blood microscopic images using supervised classifier

K. Gu et al.

Learning a no-reference quality assessment model of enhanced images with big data

IEEE Trans. Neural Networks Learn. Syst.

(2018)

A. Gupta et al.

Classification of normal vs malignant cells in b-all white blood cancer microscopic images

K. He et al.

Deep residual learning for image recognition

G. Huang et al.

Densely connected convolutional networks

M.E. Karar et al.

Intelligent medical iot-enabled automated microscopic image diagnosis of acute blood cancers

Sensors

(2022)

E. Khosla et al.

Phase classification of chronic myeloid leukemia using convolution neural networks

Cited by (16)

CAD system for intelligent grading of COVID-19 severity with green computing and low carbon footprint analysis
2023, Expert Systems with Applications
The Coronavirus Disease (COVID-19) caused a lot of mortality. The high mortality rate occurred because of the physicians’ wrong or late identification of COVID-19 severity. So, developing Computer-Aided Design (CAD) systems using Artificial Intelligence (AI) techniques is critical to help physicians correctly identify the severity of COVID-19 in the early stages of the pandemic and then decrease the COVID-19 mortality percentage. In this paper, we develop a new green CAD system using a new hybrid handcrafted feature extraction algorithm and two-stage neural network architecture to grade the COVID-19 patient based on Computed Tomography (CT) scan images as having a moderate, severe, or critical infection. Because the proposed system uses handcrafted feature extraction algorithms, it consumes minimum resources and time than recent works. The proposed system consists of three phases: lesion segmentation, feature extraction, and diagnosis. Firstly, lesions from the CT scan image are manually segmented, and then three schemes are applied to extract salient features from the segmented lesions. These schemes are the Histogram of Oriented Gradients (HOG), Speeded Up Robust Features (SURF), and a new hybrid method that consists of cascading Discrete Wavelet Transform (DWT) and Gray-Level Co-Occurrence Matrix (GLCM). Then, the Cumulative Distribution Function (CDF) is computed for each scheme to extract the statistical markers. In the grading phase, a two-stage neural network approach is used. First, the extracted features are individually trained and tested for each scheme in the first neural network stage, and then the results of the first stage are combined to train and test each patient in the second neural network stage. The performance of the proposed system was assessed on a CT image dataset of 300 COVID-19-positive patients collected from the Cancer Imaging Archive (TCIA) website. The experimental results showed that our proposed system achieved 100% accuracy and kappa when the dataset was partitioned into 80% for training and 20% for testing. Also, it achieved 95.67% $\pm$ 0.47, 99.33% $\pm$ 0.77, and 100% $\pm$ 0 accuracies and 93.48% $\pm$ 0.74, 98.997% $\pm$ 1.16, and 100% $\pm$ 0 kappa values when the data was organized using 2, 4, and 10 folds, respectively. A green complexity algorithm analysis shows that this proposed system takes O(n) time complexity and 1 h and 20 min to train and test all cases. The performed green complexity analysis shows that the proposed system consumes 117.80 g Carbon Dioxide Equivalent (CO2e), 130.80 Wh, and 0.13 tree months for the carbon footprint, the energy needed, and the carbon sequestration, respectively. These results show that the proposed work consumes fewer resources and provides a green CAD system.
γ-polyglutamic acid fermentation monitoring with ATR-FTIR spectroscopy based on a shallow convolutional neural network combined with data augmentation and attention module
2023, Chemometrics and Intelligent Laboratory Systems
The potential of attenuated total reflection Fourier transform infrared (ATR-FTIR) spectroscopy to online real-time monitor the γ-polyglutamic acid (γ-PGA) fermentation process by measuring the concentration of two key components, glucose and sodium glutamate, was investigated. Partial least squares regression (PLSR) and convolutional neural network (CNN) were selected as the multivariate calibration model to predict fermentation parameters. To solve the small sample training problem with CNN, the data augmentation strategy was firstly combined with a shallow CNN to generate a new CNN model termed as DA-CNN. To improve the performance of DA-CNN further, squeeze-and-excitation (SE) module of the attention mechanism was integrated into DA-CNN to form a DA-CNN variant termed as DA-SE-CNN, which could capture the different importance of extracted channel-wise feature to improve feature representation ability. Experiments were conducted on the γ-PGA dataset, and the results showed that the nonlinear models (CNN, DA-CNN and DA-SE-CNN) performed better than the linear method (PLSR) in all cases except the case of sodium glutamate prediction with CNN. Additionally, both DA-CNN and DA-SE-CNN perform better than CNN in all cases with limited training data. Especially DA-SE-CNN gave the best result with excellent prediction accuracy, which indicated that ATR-FTIR combined with nonlinear regression tool (DA-SE-CNN) as a rapid method to monitor γ-PGA fermentation process is feasible.
Efficient improvement of classification accuracy via selective test-time augmentation
2023, Information Sciences
Test-time augmentation (TTA) is typically used in image classification tasks to improve the classification accuracy of convolutional neural networks (CNNs). In the inference phase, TTA aggregates the predictions of a CNN over multiple random transformations of a query image to obtain the final prediction. However, TTA incurs a high computational cost owing to its effectiveness. In this study, we propose a selective-TTA method for reducing computational costs while maintaining the high classification accuracy. Given a query image, we quantify the predictive uncertainty of the CNN on the image to determine whether TTA should be performed. We perform TTA to obtain an enhanced prediction only when the uncertainty is high. Experimental results on benchmark datasets demonstrated that the proposed method significantly reduced average computational costs while attaining comparable or higher classification accuracy than conventional full-TTA. When the proportion of performing TTA was set to 10%, the computational cost was reduced by 90% with no performance degradation on most datasets.
A supervised data augmentation strategy based on random combinations of key features
2023, Information Sciences
Data augmentation strategies have always been important in machine learning techniques and play a unique role in model performance optimization processes. Therefore, in recent years, these techniques have become popular in the artificial intelligence field. In this paper, a new data augmentation strategy is proposed based on the interpretation algorithm of deep convolutional neural networks, i.e., constructing new training samples by deeply exploiting key features extracted from interpretable networks to achieve sample augmentation.
Thus, a novel supervised data augmentation approach known as Supervised Data Augmentation–Key Feature Extraction (SDA-KFE) was proposed. By introducing the Neural Network Interpreter-Segmentation Recognition and Interpretation (NNI-SRI) algorithm, an augmentation strategy is proposed that can balance the high accuracy and high robustness of the final model while ensuring a large amount of data augmentation.
The advantages of the SDA-KFE algorithm are mainly reflected in the following aspects. First, it is easy to implement. This algorithm is implemented based on the lightweight NNI-SRI algorithm, which lays the foundation for the implementation of SDA-KFE so that it can be easily implemented on convolutional neural networks. Second, this model, which is widely applicable, can be applied to almost any deep convolutional network. Through research and experiments on this proposed algorithm, SDA-KFE can be applied in graphical image binary classification and multiclassification models. Third, SDA-KFE can rapidly construct data samples with diverse variations. Under the premise of determining the classification labels of the generated samples, the distribution of the feature unit composition of the samples can be controlled. Compared with traditional data augmentation methods, SDA-KFE can control the direction of the model performance, i.e., the balance between the pursuit of high accuracy and robust performance of the model. Therefore, the novel supervised augmentation approach proposed in this paper is relevant for optimizing deep convolutional neural networks, solving model overfitting, augmenting data types, etc. The data augmentation algorithm proposed in this paper can be regarded as a useful supplement to traditional data augmentation methods, such as horizontal or vertical image flipping, cropping, color transformation, extension and rotation.
ODRNN: Optimized Deep Recurrent Neural Networks for Automatic Detection of Leukaemia
2024, Research Square
Leukemia Classification Using EfficientNetB5: A Deep Learning Approach
2024, Proceedings of the 2024 Conference of Young Researchers in Electrical and Electronic Engineering, ElCon 2024

View all citing articles on Scopus

View full text